Back to Lessons Reading Results Tables 0 pts Module 1 · Lesson 4
Introduction

Reading Results Tables

Extract what matters without getting lost

The Problem

You open a paper. Table 1 has 47 rows. Table 2 has confidence intervals, p-values, and adjusted odds ratios. Table 3 is a forest plot. Your eyes glaze over. You skip to the conclusion.

What You'll Learn

Tables contain the actual evidence. The abstract tells you what the authors want you to think. The tables tell you what they actually found. By the end of this lesson, you'll know exactly where to look and what to look for.

Table 1: Baseline Characteristics

Table 1 tells you who was in the study and whether the groups were comparable at the start.

1

What It Shows

Demographics (age, sex), comorbidities, disease severity, and other characteristics—broken down by treatment group.

2

Why It Matters

If groups are different at baseline, outcome differences might be due to those imbalances, not the treatment. This is the foundation for judging internal validity.

Characteristic Treatment (n=150) Control (n=148) p-value
Age, years (mean ± SD) 67.2 ± 8.4 66.8 ± 9.1 0.71
Male sex, n (%) 98 (65%) 94 (64%) 0.78
Diabetes, n (%) 89 (59%) 62 (42%) 0.003
Current smoker, n (%) 42 (28%) 38 (26%) 0.67
Prior bypass, n (%) 23 (15%) 19 (13%) 0.54
Spot the problem: 59% vs 42% diabetes rate with p=0.003. The treatment group has significantly more diabetics. If outcomes are worse in the treatment group, is it the treatment or the diabetes?

In a well-randomized RCT, baseline p-values should generally be non-significant. Multiple significant differences suggest randomization failure or small sample size.

Reading Table 1: The Checklist

1

Sample Size

Check n per group. Small numbers (<30 per group) mean wide confidence intervals and unstable estimates. Also look for dropouts—if n in Table 1 doesn't match n in outcome tables, people were lost.

2

Clinical Imbalances

Ignore p-values momentarily. Ask: are there differences that clinically matter? A 17% difference in diabetes (like above) is huge, regardless of whether it reaches statistical significance.

3

Missing Important Variables

What's NOT in the table? For a PAD study, you'd want ABI, Rutherford class, wound status. If disease severity isn't reported, groups may be incomparable.

4

Generalizability

Does this population match your patients? Mean age 45 in a study about claudication? That's not your typical vascular patient. Results may not apply.

Table 1 Red Flags

• Multiple significant baseline differences in an RCT
• Key prognostic variables missing
• Huge standard deviations (suggests outliers or data problems)
• Percentages that don't add up to 100%

Outcome Tables: Where the Answer Lives

Outcome tables show the primary and secondary endpoints. This is the actual evidence.

1

Find the Primary Endpoint First

The primary endpoint is what the study was designed to test. It should match the sample size calculation in the methods. Everything else is exploratory.

2

Look at Absolute Numbers

Before looking at relative risk or odds ratios, find the raw event counts. "50% reduction" means nothing without knowing if it's 10% to 5% or 0.2% to 0.1%.

Outcome Treatment
n/N (%)		 Control
n/N (%)		 RR (95% CI) p-value
Primary: Major amputation 12/150 (8%) 24/148 (16%) 0.49 (0.26-0.95) 0.03
Death 8/150 (5%) 6/148 (4%) 1.31 (0.47-3.68) 0.60
Wound healing 67/150 (45%) 58/148 (39%) 1.14 (0.87-1.49) 0.34
Reintervention 34/150 (23%) 29/148 (20%) 1.16 (0.74-1.80) 0.52
Reading this table: Primary endpoint (amputation) shows benefit: 8% vs 16%, RR 0.49, p=0.03. But secondary endpoints show no difference or even trends toward harm (death RR 1.31). The CI for the primary endpoint is wide (0.26-0.95)—barely significant.

Key Metrics to Extract

Absolute Risk Reduction Control rate minus treatment rate. In the table above: 16% - 8% = 8% ARR
Relative Risk (RR) Treatment rate / control rate. Values <1 favor treatment. Above: 0.49 means 51% relative reduction
Number Needed to Treat 1 / ARR. Here: 1/0.08 = 12.5. Treat 13 patients to prevent one amputation
95% Confidence Interval Range of plausible true effects. Narrower = more precise. If it crosses 1 (for RR) or 0 (for differences), not significant
!

The CI matters more than the p-value. A p-value of 0.03 with a CI of 0.26-0.95 tells you the effect could be anywhere from 74% reduction to 5% reduction. That's a huge range of clinical meaning.

Always calculate NNT yourself. It translates statistics into clinical decision-making: "How many patients do I need to treat to help one?"

Spotting Table Tricks

1

Composite Endpoints

"MACE (death, MI, stroke, or revascularization)" lumps major events with minor ones. If the composite is significant but driven entirely by revascularization, it's not the same as reducing death.

2

Per-Protocol vs Intention-to-Treat

ITT includes everyone randomized. Per-protocol excludes dropouts and non-compliers. Per-protocol inflates effects because non-responders are removed. Always prioritize ITT.

3

Relative vs Absolute

"50% reduction!" sounds impressive until you realize it's 0.2% to 0.1%. NNT = 1000. Always find the absolute numbers.

4

Changing Denominators

Watch for n changing between tables. If 150 patients are in Table 1 but outcomes are reported for 120, where did 30 go? Lost to follow-up is often not random.

Questions to Ask

• What's driving the composite endpoint?
• Is this ITT or per-protocol?
• What are the absolute event rates?
• Why did the sample size shrink?

Exercise: Read the Table

Answer questions about the data presented in each scenario.

Question 1 of 8

The Bottom Line

0
Total Points Earned
Exercise (0/8 correct) +0 pts
Lesson Completed +100 pts
  • Table 1 shows baseline characteristics. Check for imbalances that could confound results, and whether the population matches your patients.
  • Outcome tables contain the evidence. Find the primary endpoint first, then look at absolute numbers before relative measures.
  • Calculate NNT to translate statistics into clinical terms: how many patients must be treated to help one?
  • Trust confidence intervals over p-values. The CI tells you the range of plausible effects—that's what matters for clinical decisions.
  • Watch for tricks: composite endpoints, per-protocol analyses, relative risk without absolute numbers, and shrinking sample sizes.

Your New Skill

You can now extract the key information from results tables without getting lost in the numbers. The abstract tells you what the authors want you to believe. The tables tell you whether you should believe it.